- This implements sparse matrix-vector multiply-add both with and without transposition of the matrix.
- yax() multiplies an MxN matrix (a) against an Nx1 vector (x) and adds the result to an Mx1 vector (y).
- yatx() is the same, except it takes as input an NxM matrix (a) and multiplies its implicit transpose.
- Both vectors are dense (every element is explicit) but the matrix representation is row-major sparse.
- Each nonzero matrix element has an offset i: vals[i] is the element's value and cols[i] is its column.
- The nonzero elements of matrix row r have offsets i such that offs[r] <= i < offs[r+1].

To receive a hint, submit unfixed code.